GMGalaxies is a research programme investigating the relationship between the history of cosmic structures and their properties which can be measured from images in telescopes. For example, what happened in the history of some galaxies to transform them into passive ellipticals while others, seemingly of the same mass and in the same environment, are star-forming spirals? Even such a basic question about the link between morphology and star formation has not yet been answered, revealing our theories of galaxy formation are inadequate. This is a major concern in an era where understanding the shapes of galaxies and how they relate to the underlying dark matter is essential for progress in precision cosmology.
Current research in this area rightly gives significant attention to the crucial problem of how feedback — energy input from supernovae, active galactic nuclei, and more — affect observable properties. But as well as investigating this avenue, the GMGalaxies team have pioneered and now continue to develop and apply a new technique (“genetic modification”) to investigate systematically the role of a galaxy’s merging and accretion history at high resolution, and to enhance our understanding of large scale structure in the Universe.
The genetic modification technique involves generating multiple, slightly different sets of early-Universe conditions from which a given galaxy, halo, or void will emerge. As each version of the Universe is evolved in its own computer simulation, the initial differences lead to contrasting evolutions – for instance, the galaxy might be formed earlier or later in the Universe's history, or undergo a different number of mergers with other galaxies. All this takes place in a fully cosmological setting, replicating the accretion of gas and dark matter along filaments.
The team has now published a large number of papers applying this technique, and the code has recently been made available to replicate and build on our studies.
Showing in reverse chronological order
The central densities of dark matter in dwarf galaxies can be observationally estimated, and then compared with predictions from numerical simulations. This comparison should provide evidence for (or against) exotic physics like self-interacting or fuzzy dark matter. However, for an accurate comparison, simulations must capture the effect of feedback from stellar populations, which can gravitationally move dark matter away from its expected orbits. Previous works have established that the total mass of stars formed in a galaxy plays a dominant role in determining whether the feedback can reduce observed densities of dark matter. Here, we use high-resolution dwarf galaxy simulations with a variety of formation histories to show that the time over which the stars formed also play a significant role. Only sustained star formation over the age of the universe leads to the central densities of dark matter being reduced. We look at the observational implications of this for the search for dark matter.
Galaxies are known to assemble over time through the mergers of smaller galaxies. But how relibably the effect of any particular merger can be predicted, and for how long any effects last, remains an open question. Here we explore how mergers impact observable properties of Milky-Way-mass galaxies, using two sets of simulations conducted with different galaxy formation codes (IllustrisTNG and Wintergatan). In each set, we start from the same initial conditions for a fiducial galaxy alongside four genetically modified versions which alter the strength of an early (z=2) merger. By comparing the response of the two codes to these modifications, we explore the interplay of internal feedback mechanisms with the merging history. In both cases, a stronger merger at early times results in smaller galaxies today, nearly 10 Gyrs later; however, the strength of this trend is small compared to a much larger size difference between the two codes. Despite the difference in size normalization, both simulation suites lie on the observed size-mass relation for their respective morphological types. We discuss how it is possible to reconcile these results, highlighting both the strengths (agreement in trends) and weaknesses (disagreement in normalization) of the modern understanding of galaxy formation.
The density profiles of dark matter halos can reveal important information about the nature of dark matter. However, the inner regions of simulated profiles suffer from high levels of noise due to the insufficient number of particles at small radii. Here we present a new method to improve the accuracy of density profiles from simulated halos. Each particle in a simulation snapshot is ‘smeared’ over its orbit to obtain a dynamical density profile. This new approach greatly improves the precision of dark matter profiles at small radii, with results matching the central densities measured from higher resolution runs, all for minimal computational cost.
There is considerable interest in using empty tracts of space, known as voids, as a way to tell us more about cosmology. But defining voids rigorously remains a topic in its infancy. Here, we adopt a physically-motivated definition of voids as ‘anti-halos’ (Pontzen et al. 2016) – the precise opposite of the dense halos that host galaxies and galaxy clusters. In other words, by swapping over-dense and under-dense regions in the early Universe, voids can be transformed into halos, and vice-versa. We use posterior resimulations based on reconstructions of the nearby universe using the ‘BORG’ algorithm (Bayesian origin Reconstruction from Galaxies), and show that the anti-halo voids we obtain are consistent with expectations based on the standard cosmological paradigm. The resulting catalogue is available on Zenodo.
Galaxies like the Milky Way are surrounded by several smaller 'satellite' galaxies as a natural consequence of structure formation in the Universe. These satellites are acquired both individually and through mergers with other galaxies hosting their own satellites. The abundance and properties of such satellite galaxies are likely to be linked to the precise history of the mergers of the host galaxy and as such, might preserve some memory of these merger events. Using the VINTERGATAN-GM suite of high-resolution hydrodynamical simulations, which genetically alters a z=2 merger experienced by a MW-mass host galaxy, we study how such alterations impact the abundance of satellites around it. The changes to the target merger are inevitably compensated for by other mergers, which do so not only through the mass of the merging galaxy, but through the number of satellites they bring with them. The overall impact is that when the target merger is made smaller, the galaxy hosts more satellites for ~4.5 Gyr after the end of the merger and vice versa. Thus, the recent merger history of a host galaxy is an important source of scatter in determining its satellite abundance and conversely, in recovering the host's dynamical properties from its population of satellites
The fates of massive galaxies are tied to how massive their central supermassive black holes grow. Cosmological simulations indicate that this in turn strongly correlates with the assembly time and binding energy of the host dark matter halo — but numerical experiments also show that galaxy merger events have a strong influence. So, which is it that determines a black hole mass and the fate of its parent galaxy: the assembly time, the dark matter binding energy, or the merger history? Here, we find that in the absence of mergers, the assembly time of the dark matter halo actually has little influence on the central black hole. This lack of connection is the result of a co-rotating galaxy disc decoupling the growth of the black hole from the host halo’s properties; a merger can disrupt this disc and stimulate black hole growth. This process could be essential for establishing diversity in the properties of black holes and galaxies. We also find that a merger can increase the binding energy of the halo, explaining (at least in part) why binding energy correlates with black hole mass.
We know that galaxies across the universe have assembled from smaller components through a series of mergers. But when did the Milky Way, our own galaxy, experience its last major merger? Observations with the Gaia satellite have revealed a population of stars on highly radial orbits, plunging through the centre; this has been used to argue that a major merger took place around 10 billion years ago, at a redshift z=2. Here, we perform some of the highest resolution simulations of a Milky Way-like galaxy to date, genetically modifying a z=2 merger to have stellar mass ratios between 1:25 (minor) to 1:2 (major). Despite this huge range, we always find a similar population of radially-biased stars! We dig into the reasons for this surprising result, and argue that Gaia data needs very careful interpretation if it is to reveal the true formation history of our home galaxy.
Angular momentum is a fundamental property of galaxies that drives their morphology and orientation. In this paper, we shed light on the relation between galaxies' large scale environment and their angular momentum: we modify the initial conditions of three galaxies so as to increase or decrease how much angular momentum they will accrete 3 billion years later. We find that this causes a change in the trajectory of their infalling satellites, which subsequently spin up or down the galaxy. We show how the resulting change in stellar angular momentum changes the galaxy's size and shape.
The density profiles of dark matter halos have a remarkably self-similar form across halos of very different masses and across a large variety of cosmological models. Due to the lack of consensus on a theoretical explanation for the origin of these ‘universal’ density profiles, they are typically modeled using empirical fitting formulae. Here, we design a neural network model that is able to reproduce the known variations encapsulated by previous empirical approaches. The network actually goes further and discovers an additional factor of variation in the outer profile, which we identify as related to infall of dark matter (also known as the ‘splashback’ effect). This work makes progress toward a broader goal of extracting knowledge from neural networks about the underlying physics of cosmological structure formation.
In the latest cosmological simulations, AGN feedback is able to transform the baryon cycle around Milky Way-mass galaxies, leading to quenching. Here we study what happens when this transformation is initiated by the disruptive influence of a galaxy merger. We simulate the evolution of a star-forming disc galaxy with the EAGLE model, and use genetic modification to increase or reduce the stellar mass ratio of an individual merger. Enhancing the merger initiates AGN feedback that ejects gas from the circumgalactic medium (CGM) and prevents the galaxy’s remaining gas reservoir from being refuelled. As a result, the galaxy quenches several gigayears after the merger occurs. This kind of connection over long timescales is hard to spot in a population, but is unambiguous when studied with our genetically modified simulations.
In the near future, the sky will be surveyed with great depth by new radio and optical telescopes. This offers the opportunity to understand the relationship between cool gas (from the neutral hydrogen's radio 21cm emission) and stars in faint dwarf galaxies. Here, we use genetically modified simulations to set an expectation for the new observations: we expect that galaxies with small stellar masses will have cool gas masses that vary by orders of magnitude. Moreover, radio emission from the cool gas may be significantly offset with respect to the optical emission from stars. We argue that all this is a generic prediction of cold dark matter cosmology combined with feedback from ionising radiation and supernova feedback, and therefore constitutes an excellent forthcoming observational test of the prevailing galaxy formation paradigm.
Much of what we know about the current cosmological model relies on statistical analysis of large scale structure — the distribution of matter and galaxies in the Universe on large scales. However, individual galaxies and clusters also carry, in principle, powerful information about their cosmic origins. Here, we focus on the “local super-volume”, the region of the Universe within approximately 650 million light years of the Earth. It contains several large clusters of galaxies that may be too massive to have arisen by chance within our standard model. By quantifying this likelihood, we demonstrate that these structures can strongly discriminate between different cosmological scenarios. However, currently, the observational determination of the cluster masses is too uncertain to enable such a test. This motivates concerted efforts to improve observational mass estimates.
Cadiou et al. (2021) The causal effect of environment on halo mass and concentration, MNRAS, 508, 1, p. 1189 (full text)
Understanding the impact of environment on the formation and evolution of dark matter halos and galaxies is a crucial open problem. Studying statistical correlations in large simulations sheds some light on these impacts, but the causal effect of an environment on individual objects is harder to pinpoint. In this paper, we introduce a method for resimulating the same halo in different large-scale environments. The method involves ‘splicing’ the halo's initial conditions into arbitrary environments. As a first example, we employ the technique to show that the mass of halos is set by the density structure in their Lagrangian patch; on the contrary, the dark matter halo concentration is much more sensitive to environmental effects. We then outline how this technique will be used in future to study the interaction between galaxies and their environment.
Orkney et al (2021) EDGE: Two routes to dark matter core formation in ultra-faint dwarfs, MNRAS, 504, 3, p. 3509 (full text)
A longstanding challenge to the dark matter paradigm is the cusp-core problem, in which the central density slopes in galaxies are shallower than dark matter simulations would predict. The introduction of a centrally fluctuating gravitational potential, typically taking the form of gas flows driven by bursty star formation, has been shown to address this challenge by reducing the central density in galaxy formation simulations. However, it remains unclear whether this mechanism can explain the shallow central densities of those dwarf galaxies where star formation is least efficient. This paper investigates the cusp-core problem in the lowest mass dwarf galaxies with a suite of cosmological zoom-in and genetically modified simulations. We find that a centrally fluctuating gravitational potential can lower the central density of galaxies from within the half light radius by up to a factor 2. We show that these fluctuations owe not only to gas flows, but also to impulsive heating from minor mergers. This provides a mechanism by which ultra-faint dwarf galaxies can have their central density lowered over time, even after their star formation is quenched at high redshift.
Cadiou, Pontzen, and Peiris (2021) Angular momentum evolution can be predicted from cosmological initial conditions, MNRAS, 4, 5480 (full text)
This paper investigates angular momentum growth in the universe, and in particular whether it is chaotic or can be predicted accurately from the initial conditions. Understanding the angular momentum of dark matter haloes and galaxies is crucial for improving the accuracy of weak lensing studies, which need to disentangle the intrinsic orientation of galactic disks from the gravitational distortion of the image. Our study involves implementing an algorithm to apply genetic modifications of the angular momentum content of the initial conditions. We apply such modifications to seven halos while keeping their initial mean density and large-scale environment fixed. This allows us to draw a cause-and-effect link between changes in the initial angular momentum conditions and its late-time magnitude. We find that the angular momentum can be predicted accurately from the reference simulation, with a much better accuracy than expected from tidal torque theory. We conclude that angular momentum of a galaxy can be determined from the initial conditions, provided one can accurately predict which region of gas will infall.
Lucie-Smith et al (2020) Deep learning insights into cosmological structure formation, arXiv:2011.10577 (full text)
This paper presents insights into the formation of dark matter halos using a deep learning framework. While the non-linear evolution of matter can be computed using cosmological simulations, a theoretical understanding of this complex process remains elusive. We trained a three-dimensional convolutional neural network (CNN) to learn the non-linear relationship between the initial density field and the final mass of dark matter halos. The key aspect of our work lies in developing techniques that allow us to physically interpret the learnt mapping. We removed anistropic information about the initial density field from the inputs and re-trained the CNN. We found no change in the model’s predictive accuracy, meaning that the features learnt by the CNN are equivalent to spherical averages over the initial density field. Our work shows that interpretable deep learning frameworks can provide a powerful tool for extracting insights into cosmological structure formation.
Sanchez et al. (2021) One-Two Quench: A Double Minor Merger Scenario, ApJ, 2, 116 (full text)
This paper investigates the suppression of star formation, or “quenching,” in a suite of genetically modified simulations. Beginning with a Milky Way-mass galaxy, we modify the mass of one dwarf satellite, performing three additional simulations. Each of these subsequent galaxies have fixed large scale structure and final main halo masses; however, the modification of the LMC satellite results in differences to the halo accretion history, including changes in the sequence and timing of other dwarf satellite mergers with the main galaxy. Furthermore, these differences result in widely varying galactic evolution. Of the four galaxies, two remain star forming disks similar to our Milky Way, while two of them quench fully. The conclusion of the paper is that these subtle changes in the accretion history — as probed by the genetic modification — can drive major differences in the evolution of Milky Way mass galaxies.
Pontzen et al. (2021) EDGE: a new approach to suppressing numerical diffusion in adaptive mesh simulations of galaxy formation, MNRAS, 2, 1755 (full text)
This paper outlines a completely new use for genetic modification: improving the numerical accuracy of our galaxy formation simulations. “Adaptive mesh refinement” is a particular approach to simulations which results in very accurate treatment of features such as shocks and instabilities, and which we therefore use for studying dwarf galaxies. However, when the entire galaxy moves at high speeds across the simulation, the accuracy degrades. This is especially a problem when the speed of the galaxy is large compared to its own internal velocity dispersions (which is the case for small galaxies). Genetic modification gives us a neat solution: we can demand that the speed of the galaxy relative to the simulation rest frame is minimal. The paper shows that such an approach leads to even more accurate gas dynamics, and so better insight into small galaxies at high redshift.
Stopyra, Peiris, and Pontzen (2021) How to build a catalogue of linearly evolving cosmic voids, MNRAS, 3, 4173 (full text)
This paper studies voids: large and mostly empty regions that make up most of the volume of the Universe. Our goal is to understand how the formation and evolution of these voids can be understood using simple ‘linear’ models. Normally, understanding structure formation requires computationally intensive, non-linear numerical calculations. Linear models, by contrast, are simple to understand and quick to evaluate. The Zel’dovich approximation is one such linear model, and in this paper, we quantified how well it describes the density of matter as a function of distance from the centre of the void. We found that it worked well for voids with a radius larger than 5 Mpc (approximately 16 million light years). We also showed that these voids can be well-described as ‘anti-halos’ (Pontzen et al. 2016) – the precise opposite of the dense halos that host galaxies and galaxy clusters. In other words, by swapping over-dense and under-dense regions in the early Universe, voids can be transformed into halos, and vice-versa. This allows us to directly link void formation to the state of the early universe, which will help us to better understand how the universe came to be the way it is today.
Davies, Crain, and Pontzen (2021) Quenching and morphological evolution due to circumgalactic gas expulsion in a simulated galaxy with a controlled assembly history, MNRAS, 1, 236 (full text)
This paper investigates the influence of a dark matter halo’s assembly history on the properties of its central galaxy and circumgalactic medium (CGM) in the EAGLE galaxy formation model. By employing the genetic modification technique, we can take a present-day halo hosting a star-forming, Milky Way-like disc galaxy and systematically accelerate or delay its assembly. We find that shifting the halo assembly to earlier times yields a spheroidal, quenched galaxy, while shifting to later times yields a more actively star-forming disc. This occurs because the halo assembly history modulates the ejection of gas from the CGM by AGN feedback, and thus modifies how readily the CGM can cool and fuel star formation in the central galaxy. Genetic modification allows us to forge a causal connection between these processes, as we only adjust the assembly history. These results demonstrate that the ejection of gas from the CGM is a crucial, previously under-appreciated step in galaxy quenching.
The primary code output from the GMGalaxies project is our initial conditions generator, genetIC, which is designed to create initial conditions for N-body and hydrodynamical zoom simulations which can be tweaked or ‘genetically modified’. The purpose is to make fine custom adjustments to the history and environment of a galaxy, and so enable its dependence on these factors to be investigated systematically.
GenetIC accomplishes this by generating random initial conditions and then allowing the user to specify the required variations. Changes can be made in a large variety of linear variables including, for example, the average dark matter overdensity in specified regions or sub-regions of a galaxy's Lagrangian patch. By choosing the modifications carefully, a selection of galaxies with different accretion histories can be efficiently ‘scanned’ using a relatively small number of simulations.
Assuming that you are interested in applying this technique, note that the final changes to a halo accretion history cannot be perfectly predicted from a given modification to the initial conditions. There is an art to guessing the best set of modifications to achieve a particular effect. It is important to verify that the modifications imposed have had the desired effect (probably using a cheap dark-matter-only simulation, before spending CPU time on hydrodynamics!). To see what is possible in practice and how to achieve it, take a look at our publications.
Download the latest release of the code and manual from the github releases page. The manual also includes instructions for running genetIC as a docker container.
Members of the GMGalaxies collaboration also maintain and develop analysis codes, pynbody and tangos.
Pynbody is an analysis package for astrophysical N-body and hydrodynamical simulations, supporting Python 3.5 and later. It enables users to analyse their simulations without worrying about file formats. The code has been used in a variety of astrophysical domains, ranging from cosmology to star and planet formation, for the last decade.
Tangos builds on the foundation of pynbody (or equivalent analysis packages such as yt) to create rich, interactive databases summarising the results of a cosmological simulation. It is particularly crucial for the GMGalaxies team because it allows us to link information about the development of galaxies over cosmic time and across simulations with different ‘genotypes’.
For more information about using these codes, please visit their github pages.
Recently we have also added a real-time GPU-based visualization code, topsy. Topsy uses pynbody to load files, then performs SPH rendering on the GPU to enable rapid scientific exploration.